Pertussis Diagnosed With Voice-Recognition Technology

Daniel M. Keller, PhD

October 30, 2012

SAN DIEGO, California — The paroxysmal phase of a pertussis cough is so distinctive that, using voice-recognition technology, researchers have developed a system that can accurately classify a case, or at least assign a probability of it being pertussis, after just a few seconds of coughing.

The technology can accurately distinguish pertussis coughs from croup and nonpertussis coughs, Philip Polgreen, MD, MPH, associate professor of internal medicine at the University of Iowa in Iowa City, and director of the Infectious Diseases Society of America's Emerging Infections Network, explained here at ID Week 2012.

Young children in particular develop a paroxysmal phase as the cough becomes more severe, resulting in bursts of rapid coughs followed by gasps and the characteristic whooping sound. Older children and adults can lack the paroxysms and whoops.

Dr. Polgreen and some acoustical engineers collected sound files from YouTube and public-health Web sites of pertussis coughs, as well as sound files from coughing children with nonpertussis conditions, including croup, bronchitis, and asthma.

Using the algorithms that underpin all voice-recognition software, the researchers used the frequency characteristics of the sounds to produce logarithmic representations that model the way human beings perceive sound. The resulting spectrograms displayed frequency in relation to time.

Spectrograms Decidedly Different

Spectrograms from typical nonpertussis and pertussis recordings were quite different. The nonpertussis recording was smooth and fairly uniform, whereas the pertussis recording showed the sharp, rhythmic spikes of the cough followed by quieter inhalation times. The rhythm is as distinctive as the cough sound, which helps to classify the spectrogram, according to the researchers.

"So often early cases are missed in an outbreak. If physicians have seen classic cases they can often recognize it, but if you don't think about pertussis and haven't heard it, even the classic cases can be missed," Dr. Polgreen told Medscape Medical News. "We were able to correctly classify the different coughs with a sensitivity and specificity of over 90% each using a neural-network-based prediction," he explained.

With their small dataset of 16 nonpertussis sounds and 31 pertussis coughs, the researchers correctly classified 92% of pertussis coughs and incorrectly classified 8% of pertussis coughs as nonpertussis. Nonpertussis coughs were correctly classified 93% of the time and incorrectly classified 7% of the time. Gathering and analyzing more data should allow the system to become more sensitive and specific.

Many Ways to Apply the Technology

The developers were able to "build a robust classifier to identify possible pertussis cases with classic symptoms in the paroxysmal phase" using small sample of coughs. This diagnostic aid could be made into a computer, smart phone, tablet, or cloud-based application (app) that can be used by anyone who can create a high-fidelity sound file. As an app, it could help parents determine when to seek medical attention.

The researchers also envision using the technology to build "cough-based" surveillance systems. Because the project was completed using existing Web-based cough files, crowd-based surveillance approaches should be feasible.

Dr. Polgreen sees the technology as "a great public-health application" that might be appropriate for analyzing other forms of coughs and diseases. "But right now, with the pertussis outbreak and reemergence of pertussis, we'd like to focus on pertussis and building a tool that we can...distribute," he said.

Jeffrey Kahn, MD, PhD, professor and chair of pediatric infectious diseases at the University of Texas Southwestern Medical Center in Dallas, who was not involved in the study, told Medscape Medical News that the results are provocative but need to be verified. "If this is true and they perfect this could be very, very useful," he said.

"I think one of the things that we're faced with in all areas of infectious diseases is adequate screening tests.... They have to be sensitive and specific, they have to be reliable, and they have to be quick," Dr. Kahn said. He added that it might be possible to incorporate other diagnostic techniques into this technology, such as polymerase chain reaction (PCR), and improve its positive predictive value.

Dr. Kahn sees this technology being particularly useful in infants and young children who are partially immunized and typically do not present with the classic whooping cough.

"Presumably, this is a reflection of where in the airway the disease is occurring," he explained. "I guess you could theoretically expand this to other pathogens that affect the upper airway and begin to make a diagnostic tree based on the frequency of cough. This would be an interesting way of differentiating what we now think of as reactive airway disease from viral infections or even bacterial infections, including pertussis."

Data Confirm Long-Term Suspicion

Andrew Pavia, MD, chief of pediatric infectious diseases at the University of Utah in Salt Lake City, who was also not involved in the study, called the technology "a fascinating idea." He told Medscape Medical News that some physicians have claimed for decades that there were differences in the tone of the pertussis cough, even when kids did not have the classic whoop.

The fact that an app can be used to distinguish pertussis from nonpertussis coughs is particularly exciting. "If it really works, this could be something that you walk around with on your iPhone or Android device," Dr. Pavia said.

He hopes the researchers test the software on more cases of early and late pertussis in different age groups. It might pick up pertussis before the classic whoop or paroxysms appear. In the future, when you have 1 case of PCR-proven pertussis in a family, it might be possible to use the spectrogram to quickly diagnose the rest of the family, which would save money, he said.

"We're going to see probably 10,000 cases of pertussis in the United States in the next 6 to 8 months; it's really a big problem. Testing requires PCR, which often takes a day or 2. As a bedside tool that every physician with a smart phone could use, this could be really exciting," Dr. Pavia explained. Potentially, you could record the cough at home and send the sound clip to the physician, he added.

There was no commercial funding for the study. Dr. Polgreen, Dr. Kahn, and Dr. Pavia have disclosed no relevant financial relationships.

ID Week 2012: Abstract 423. Presented October 18, 2012.